The client
A home goods retailer with about 320 stores, 3 distribution centers, and 42% of its sales online. Online orders run through Salesforce Commerce Cloud, store prices and promotions through Oracle Retail, and the teams work in Slack.
The engagement
The chief operating officer and the chief marketing officer sponsored one engagement. Operations wanted held orders and returns cleared faster, and marketing wanted promotions approved faster. Both problems led back to the promotion. Five workflows were in scope:
- Promotion setup: about 1,400 a year, each entered once for the stores and again for the website.
- Creative review: about 9,800 emails, banners, and ads a year, 4,300 of them with a price or an offer.
- Order holds: 410,000 a year on 6.8M online orders.
- Returns: 2.0M a year.
- Price adjustments and goodwill credits: 162,000 a year.
What the pilot found
Four forward-deployed engineers ran a 4-week pilot with merchandising, e-commerce, the creative studio, fraud, returns, and customer care. They studied 12 months of promotion, order, and support records, and read the policies. They also interviewed the people who do the work.
- One promotion, 3 records. 1 in 7 promotions disagreed between systems on dates, brands, or discounts. Those mismatches caused 13% of order holds and 31% of price adjustments.
- Every furniture order over $1,500 went to manual fraud review, by a rule no document described. 71% were released unchanged, after 31 hours on hold.
- Ads were checked against the brief, not the promotion as entered. In a sample of 600, 9% of sale ads showed a "was" price that broke the retailer's own pricing policy.
Marketing budgeted 7 hours to set up a promotion, about $480. Measured, it took 19 hours across 5 teams, about $1,290. Operations counted $4.10 of analyst time per held order. Counting cancellations, customer contacts, and rushed shipping, a held order cost $11.60. The redesign brings it to $4.80.
The redesign
The focus followed one promotion from the brief to the orders it held up. An agent now builds each promotion once, in the store and online systems together, and test-prices sample baskets before launch. A second agent checks every ad against the promotion, including each "was" price against the price history. For held orders, agents gather the evidence and check stock everywhere before an order is held. The policies and the approvers did not change.
Before16 steps
28 hours
After7 steps, 1 human decision
7 hours
- Manual step
- Agent step
- Person decides
Promotions with a setup mismatch
- Before
- 1 in 7
- After
- 1 in 50
- Change
- 86% fewer
Holds caused by promotion setup
- Before
- 53,300 a year
- After
- 7,500 a year
- Change
- 86% fewer
Process steps to clear a hold
- Before
- 16
- After
- 7
- Change
- 56% fewer
Average time on hold
- Before
- 28 hours
- After
- 7 hours
- Change
- 75% shorter
Holds cleared with no human correction
- Before
- 14%
- After
- 76%
- Change
- +62 pts
Held orders canceled
- Before
- 9.2%
- After
- 4.7%
- Change
- 49% fewer
| Before | After | Change | |
|---|---|---|---|
| Promotions with a setup mismatch | 1 in 7 | 1 in 50 | 86% fewer |
| Holds caused by promotion setup | 53,300 a year | 7,500 a year | 86% fewer |
| Process steps to clear a hold | 16 | 7 | 56% fewer |
| Average time on hold | 28 hours | 7 hours | 75% shorter |
| Holds cleared with no human correction | 14% | 76% | +62 pts |
| Held orders canceled | 9.2% | 4.7% | 49% fewer |
4 weeks of discovery, then 4 weeks of deployment
- Weeks 1 to 4, discovery: the pilot above maps the process, interviews the team, and sets the baseline for every figure on this page.
- Week 5, build: the agents are built on the redesigned promotion setup, the pricing policy, and the hold and returns procedures.
- Week 6, shadow: the agents build every new promotion and check every held order with read-only access, and their work is compared with the team's.
- Week 7, validate: the agents are tested on 6 months of past promotions and 3 months of holds, and the heads of pricing and fraud sign off.
- Week 8, deploy: live in Oracle Retail, Commerce Cloud, Manhattan, and Workfront, with every action logged.
- From week 8, train: the engineers stay on site with the teams whose work changed. Returns and customer care credits follow, 4 weeks each.
What the agent does
- When a merchant sends a promotion brief, the agent sets up the promotion in the store and online systems.
- It test-prices sample baskets in both, to catch errors such as discounts that stack or a brand that should be left out.
- It checks the ads and emails against the promotion as entered, drafts any fixes, and stops.
- The pricing analyst and the brand reviewer approve in Slack, and the promotion goes live on time everywhere.
Results
A promotion now takes 9 hours across 3 teams, down from 19 across 5. Ads clear review in 2 days instead of 6, and every sale price is checked against its history. Refunds outside policy fell from 5.9% to 1.8%, and a return is refunded in 4 days instead of 9. Every figure is measured against the pilot baseline.
| Source | Annual value | How |
|---|---|---|
| Capacity released | $4.7M | 121,400 hours: 100,400 in fraud, returns, and customer care at $33 fully loaded, and 21,000 in merchandising, e-commerce, and creative review at $68 |
| Refund leakage removed | $4.2M | 4.1 points of out-of-policy refunds on 2.0M returns, 82,000 refunds at a $51 average net loss |
| Promotion margin recovered | $2.5M | Margin given away on mis-set promotions down from $3.0M to $0.5M a year, as mismatches fell from 1 in 7 promotions to 1 in 50 |
| Orders kept | $1.5M | Canceled held orders down from 37,700 to 17,100 a year, 20,600 orders at $190 and a 38% gross margin |
| Credits avoided | $1.0M | Price adjustments and credits caused by promotion mismatches down from 50,200 to 7,000 a year, at $23 each |
| Total | $13.9M | Against the $60.5M a year it costs to run the workflows in scope: about 520 fraud, returns, and customer care staff at $33 and 85 merchandising, e-commerce, and creative review staff at $68, plus $12.7M of refunds outside policy, margin given away on mis-set promotions, and price adjustments and credits |
What stays with people
Agents never set a price, approve a promotion, or decide an order. The pricing analyst approves every promotion, brand and legal approve every offer, and the merchandising lead keeps the calendar and the margins. Fraud analysts decide every cancellation for fraud. Refunds outside policy, goodwill credits, and any decision to block a customer stay with people. Every agent action records the rule, the evidence, and the approval.
